Machine learning models for efficient characterization of Schottky barrier photodiode internal parameters.

Area of Science:

• Semiconductor device physics
• Materials science
• Machine learning applications

Background:

• Schottky barrier diodes (SBDs) are crucial semiconductor devices.
• Accurate extraction of internal parameters like barrier height and ideality factor is vital for device optimization.
• Traditional methods often rely on complex physical models, such as the thermionic emission (TE) equation, which can be computationally intensive.

Purpose of the Study:

• To develop and validate artificial neural network (ANN)-based models for analyzing and extracting internal parameters of Schottky photodiodes (SPDs).
• To demonstrate the capability of these models to predict device parameters without prior knowledge of the nonlinear thermionic emission (TE) equation.
• To assess the accuracy and efficiency of the proposed machine learning (ML) approach in device characterization.

Main Methods:

• Training and evaluating ML models on thirty-six datasets from graphene oxide (GO) doped p-Si SBDs.
• Utilizing current-voltage-temperature-illumination data under varying GO doping levels (0-10%) and illumination intensities (0-30 mW/cm²).
• Incorporating independently calculated parameters (barrier height, ideality factor, series resistance) into the training data and predicting parameters at higher intensities and external doping levels.

Main Results:

• ANN models achieved high accuracy with mean square error and mean absolute error below 0.003 across all datasets.
• The models successfully predicted internal parameters without explicit reliance on the TE equation.
• Accurate predictions were demonstrated on both model development data and external datasets with varying conditions.

Conclusions:

• ANN-based models effectively capture photodiode responses and accurately predict internal SBD parameters.
• The proposed ML approach offers a significant reduction in analysis time for device development cycles.
• These models show potential for broad applicability to diverse datasets in various scientific and engineering fields.